Initial Uncertainty Estimation

We propagate uncertainty through the pipeline. This module initializes the uncertainty based on the bias level, the dark level, the gain, read noise level, bitrate, and signal levels.

Concept

The initial uncertainty is characterized in level 1 data products by passing the input data file and associated metadata into a subroutine which computes and inserts these uncertainty values in place. Using CCD / instrument characteristics such as the bias level, the dark level, the gain, the read noise level, and the signal bitrate, an estimation of the noise levels can be computed.

Photon / shot noise is generated from the gain and input signal. The dark noise is generated from the dark level and a poisson distribution. The read noise is generated from a normal distribution using the provided read noise level. These individual noise terms are added in quadrature, and returned from a subroutine. The initial uncertainty estimate is then folded into the data object as a ratio of the noise level with the provided / computed data photon counts.

Applying correction

The correction is carried out primarily in the punchbowl.level1.initial_uncertainty.compute_uncertainty function:

punchbowl.level1.initial_uncertainty.compute_uncertainty(data_array: ndarray, bias_level: float = 100, dark_level: float = 55.81, gain_left: float = 4.9, gain_right: float = 4.9, read_noise_level: float = 17, bitrate_signal: int = 16) → ndarray

With an input data array compute a corresponding uncertainty array.

If you wish to incorporate this as a Prefect task in a custom pipeline, using something like the punchbowl.level1.initial_uncertainty.update_initial_uncertainty_task is recommended.